Responsible AI and your next project
A model that decides who gets a loan, who gets hired or which invoice is blocked affects real people. That's why countries have set principles. In Saudi Arabia, SDAIA (the Saudi Data and AI Authority) published AI Ethics Principles; version 2.0 has 7 principles:
| Fairness | No discrimination against a group by gender, age or region. |
| Privacy & security | Protect personal data and comply with data-protection law. |
| Humanity | Serve people and respect their values and dignity. |
| Social & environmental benefits | A positive effect on society and the environment. |
| Reliability & safety | Works consistently and safely, and is tested before launch. |
| Transparency & explainability | People know a decision came from AI, and we can explain why. |
| Accountability & responsibility | Someone is answerable for the system's decisions and their outcomes. |
Bias comes from the data. If a company trains a hiring model on past staff who mostly came from one university, the model learns to favour that university, not because it's better, just because the past looked that way. The fix starts with checking the data before training and measuring performance for each group separately.
A model needs watching after launch. The world changes: a sales model trained on 2024 can weaken when customer behaviour shifts or a new competitor arrives. This is model drift. So we track the error every month and retrain when it grows. It's the “deployment” phase of CRISP-DM, and it never really ends.
Project ideas that combine what you've learned (good for a graduation project or your CV):
| Project | Model | Connects to |
|---|---|---|
| Demand forecasting for a small shop | Moving average, then regression | SQL to prepare sales, a BI dashboard for results |
| Churn prediction for a subscription or gym | k-NN or a decision tree | A confusion matrix and a customer outreach plan |
| RFM segmentation for an online store | k-means | SQL to compute RFM, marketing advice per segment |
| Auto-routing IT support tickets | A language model with a well-designed prompt | ITIL from the governance track, and the prompt guide |